Technical articles on AI agent management, cost tracking, encryption, and building developer tools.
Four new agents (Crush, Kilo, Kiro, Droid) bring the total to 11. Plus session checkpoints, voice commands, per-message cost tracking, mobile code review, and OpenTelemetry export for Grafana, Datadog, and Honeycomb.
Shared dashboards, per-developer attribution, and team budget alerts for the Power tier.
How token-based pricing works, input vs. output cost differences, and why cache tokens matter.
Claude Channels are free and native. Styrby adds multi-agent support, E2E encryption, and cost tracking. Here is what matters.
Leave agents running, get push notifications, approve permissions from your phone, and use budget alerts as safety nets.
TweetNaCl box encryption, key exchange, and zero-knowledge architecture for code sessions.
Engineering decisions, key generation, performance characteristics, and lessons learned.
Two different approaches to controlling AI agents remotely. Architecture, encryption, and agent coverage compared.
Freelancers and agencies billing AI costs to clients. Use session tags and project paths for cost attribution, then export for invoicing.
Setting up daily, weekly, and monthly limits with three graduated actions: notify, slow down, and hard stop.
SQLite setup, queue pattern, sync on reconnect. Tutorial-style with code examples.
Real pricing for Claude, GPT-4o, Gemini Pro, and more. Per-token costs, typical session costs, and how to track them.
Why 24/7 notifications cause fatigue. Configuring time windows and critical alert bypass.
The security case for mobile approval. Risk badges, blocked tool lists, and real examples of dangerous commands.
How AI coding agents handle permissions differently, from Claude's allowlists to Codex's sandbox. Plus: when remote mobile approval makes sense.
Token bucket, sliding window, fixed window. Per-endpoint vs. per-user limits with Upstash Redis.
How encrypted replay works. Filtering by cost, agent, and project. Bookmarking sessions for later.
Before: checking each agent separately. After: unified view with live status. Practical workflow changes.
The manual way vs. automated real-time tracking with threshold alerts. A practical comparison.
Postgres vs. Firestore, row-level security vs. Firestore rules, realtime subscriptions, and the self-host option.
SQLite on mobile, IndexedDB on web, and the sync protocol that ties them together.
Threshold tuning, graduated actions, and the tradeoffs between period-based and rolling windows.
Color-coded error classification that saves debugging time by identifying the real source of failures.
Annual cost analysis, hidden costs like context window usage and retry loops, and how to estimate monthly spend.
The problem with five different terminals. Unified status, cost aggregation, and color-coded agents.
Full security architecture published for peer review. Encryption, auth, data handling, and what we would change.
Real risks vs. overhyped fears. Unauthorized file access, credential exposure, and practical security measures.
iOS app timeline, Android plans, upcoming features, and what we are deliberately not building.
Usage patterns across Claude, Codex, Gemini, and more. Why developers don't pick just one, and the overhead that creates.
Military background brings discipline to software engineering. Steel Motion LLC and the tools we build.
The problem: multiple AI agents with no unified cost or permission visibility. The solution we built.